J
shadders said:jcbrum said:Shadders, the quantisation value is exact within the constraint of the wordlength employed, This is what also determines the noise level.
If you can't hear the noise, then the result is audibly the same as an infinite wordlength.
Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.
Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems, but you can't hear digits, so you have to listen to analogue. The test is whether you can actually hear the 'noise' in the result.
That is what fr0g's test is about.
JC
Hi,
I never disputed that quantisation creates the noise in a reconstructed waveform
My statement is that Shannon-Nyquist is based on EXACT sample values.
jcbrum said:If you can't hear the noise,...
Agreed - if you can't hear the noise. There may be others who can.
jcbrum said:Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.,
I have never stated that analogue systems have infinite S/N. Analogue systems do have near infinite resolution, but this is masked by the noise.
jcbrum said:Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems
Agreed, theory does seem to always be better than any practical analogue system. This is not in dispute.
jcbrum said:The test is whether you can actually hear the 'noise' in the result
The test needs to be documented such that its validity can be ascertained and repeated. Partial process statements and partial results are not conclusive evidence.
One question - if you cannot hear an absoute -90dB signal - then why is dither implemented for 16bit words ?
Regards,
Shadders.
Jota180 said:Clare Newsome said:Alec said:Clare Newsome said:Lordy is this thread still going?!
Starting to remind me of a man I once met who never watched (or read) anything anything fictional, as he couldn't accept it was 'made up'.
He was particularly scathing about sci-fi, arguing that the 'sci' bit was invalid as there were so many breaches of scientific fact involved.
How sad.
That needs unpacking by a superior mind to mine, because I tried, and it felt like I'd just read manic's posts all over agian.
Didn't think it was that obtuse (but then I am feverish with lurgy at moment!)
Just saying it seems some people seem determined to make science - rather than enjoyment - the focus of their thinking at all times. Which is odd as we're inherently talking about entertainment and emotional connection.
The guy I met couldn't understand how I could lose myself in the narrative of, say, Star Wars, or even a James Bond movie, while knowing it wasn't real. I equally couldn't fathom how you could go through life analysing everything for its 'reality' before allowing yourself to enjoy it...
I think my main concern, coming back to the topic at hand, in the audiophile sphere is snake oil. It relies on the fact that the emotional amongst us can be carried away without realising it or even admitting to the fact they are succeptable to the oil sales pitch, to the placebo effect and to confidence tricks. Proper double blind testing can remove the chance that the reviewers will subconsciously influence their own conclusions.
I know people think, don't insult my intelligence, I'm too smart to get caught out like this but for your average, emotionally 'standard' person it's in your make up. It's why you're the way you are with emotional links to other people, links based on trust, and it's this ability to trust others that is also your weakness for confidence tricksters, for snake oil salesmen and the like.
The guy you met lacked your ability when it came to losing yourself in fiction but he also probably would not lose himself in anything including any claims made by the hifi manufacturing industry for example. Not even subconsciously.
Speaking as someone with aspergers I accept nothing at face value, from no one, without evidence, question everything and if some company is going market product A as some big improvement I want to see the empirical evidence.
I don't understand why publications like this don't conduct double blind tests in a number of areas as to my mind there's no other satisfactory way.
busb said:shadders said:jcbrum said:Shadders, the quantisation value is exact within the constraint of the wordlength employed, This is what also determines the noise level.
If you can't hear the noise, then the result is audibly the same as an infinite wordlength.
Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.
Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems, but you can't hear digits, so you have to listen to analogue. The test is whether you can actually hear the 'noise' in the result.
That is what fr0g's test is about.
JC
Hi,
I never disputed that quantisation creates the noise in a reconstructed waveform
My statement is that Shannon-Nyquist is based on EXACT sample values.
jcbrum said:If you can't hear the noise,...
Agreed - if you can't hear the noise. There may be others who can.
jcbrum said:Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.,
I have never stated that analogue systems have infinite S/N. Analogue systems do have near infinite resolution, but this is masked by the noise.
jcbrum said:Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems
Agreed, theory does seem to always be better than any practical analogue system. This is not in dispute.
jcbrum said:The test is whether you can actually hear the 'noise' in the result
The test needs to be documented such that its validity can be ascertained and repeated. Partial process statements and partial results are not conclusive evidence.
One question - if you cannot hear an absoute -90dB signal - then why is dither implemented for 16bit words ?
Regards,
Shadders.
This has been answered several times! Without dither, the artefacts would be much higher than -90dB, so high, they'd often be audible. It's the added dither that "averages out" the quantisation error but at the expense of increasing the noise-floor upwards but at a level that's inaudible. It's one of those rare win-win situations!
These links, if you haven't already read they may help:
http://www.head-fi.org/t/415361/24bit-vs-16bit-the-myth-exploded
http://www.audiomisc.co.uk/inadither/Page1.html
busb said:Without dither, the artefacts would be much higher than -90dB, so high, they'd often be audible
shadders said:busb said:shadders said:jcbrum said:Shadders, the quantisation value is exact within the constraint of the wordlength employed, This is what also determines the noise level.
If you can't hear the noise, then the result is audibly the same as an infinite wordlength.
Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.
Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems, but you can't hear digits, so you have to listen to analogue. The test is whether you can actually hear the 'noise' in the result.
That is what fr0g's test is about.
JC
Hi,
I never disputed that quantisation creates the noise in a reconstructed waveform
My statement is that Shannon-Nyquist is based on EXACT sample values.
jcbrum said:If you can't hear the noise,...
Agreed - if you can't hear the noise. There may be others who can.
jcbrum said:Analogue sound and analogue systems do not have infinite resolution and infinite SNR either.,
I have never stated that analogue systems have infinite S/N. Analogue systems do have near infinite resolution, but this is masked by the noise.
jcbrum said:Digital systems can easily have better resolution and better SNR, in mathematical terms, than practical real world analogue systems
Agreed, theory does seem to always be better than any practical analogue system. This is not in dispute.
jcbrum said:The test is whether you can actually hear the 'noise' in the result
The test needs to be documented such that its validity can be ascertained and repeated. Partial process statements and partial results are not conclusive evidence.
One question - if you cannot hear an absoute -90dB signal - then why is dither implemented for 16bit words ?
Regards,
Shadders.
This has been answered several times! Without dither, the artefacts would be much higher than -90dB, so high, they'd often be audible. It's the added dither that "averages out" the quantisation error but at the expense of increasing the noise-floor upwards but at a level that's inaudible. It's one of those rare win-win situations!
These links, if you haven't already read they may help:
http://www.head-fi.org/t/415361/24bit-vs-16bit-the-myth-exploded
http://www.audiomisc.co.uk/inadither/Page1.html
Hi,
I had a quick scan - for the head-fi - they do state "perceived" increased S/N. Perceived is the key word here - not actual increase in S/N.
For the latter audiomisc - Jim Le Surf is using a single tone at -90dB - hence only using 2 bits from the possible 16bits - and i think he is using processor gain to achieve the 120dB S/N.
I would need to read more closely - to confirm the above.
Again, valid analysis, but the context does need to be explained, without, can be misleading.
busb said:Without dither, the artefacts would be much higher than -90dB, so high, they'd often be audible
Can you explain how you have derived this statement - i was not able to locate in the text links you provided. Thanks.
Regards,
Shadders.
jcbrum said:Dither should be added to any low-amplitude or highly-periodic signal before any quantization or re-quantization process, in order to de-correlate the quantization noise from the input signal and to prevent non-linear behavior (distortion); the lesser the bit depth, the greater the dither must be. The result of the process still yields distortion, but the distortion is of a random nature so the resulting noise is, effectively, de-correlated from the intended signal. Any bit-reduction process should add dither to the waveform before the reduction is performed.
In short, dither de-correlates quantisation errors from the signal source.
JC
Exactly the problem with this thread, then.busb said:I don't share your confidence in DB ABX testing - it doesn't deal with false negatives & needs to be conducted by professionals, is time consuming & rather boring to take part in so say nothing of the expense. No one in the industry is going to risk such tests, IMO. I've bee to demos of speaker cables where I've stated tha I couldn't really tell much difference to be challenged by the manufacturer with such statements as "Are you sure!" or "Well, it's pretty obvious to the rst of us!" etc. Such tests always start at the cheap end then work upwards. I never been to a demo whee the audiance has ever been asked to guess what level a particular cable falls in their range!
altruistic.lemon said:Exactly the problem with this thread, then.busb said:I don't share your confidence in DB ABX testing - it doesn't deal with false negatives & needs to be conducted by professionals, is time consuming & rather boring to take part in so say nothing of the expense. No one in the industry is going to risk such tests, IMO. I've bee to demos of speaker cables where I've stated tha I couldn't really tell much difference to be challenged by the manufacturer with such statements as "Are you sure!" or "Well, it's pretty obvious to the rst of us!" etc. Such tests always start at the cheap end then work upwards. I never been to a demo whee the audiance has ever been asked to guess what level a particular cable falls in their range!
jcbrum said:DSD is an interesting case in point when considering quantisation errors.
Since DSD has only a one bit wordlength, one might suppose that the quantisation error is huge. In fact so large as make one assume it might be insurmountable, since it is 32,000 times larger than 16bit.
The reality is that DSD is very heavily dependent on dither and, in particular, noise shaping technology, yet produces results which are audibly undetectably different from multibit technology, such as 24bit pcm.
JC
jcbrum said:p.s. Just in case any readers don't know what DSD is, the following note might be of assistance . . .
DSD is 1-bit, has a sampling rate of 2.8224 Mhz. The output from a DSD recorder alternates between levels representing 'on' and 'off' states, and is a binary signal (called a bitstream). The long-term average of this signal is proportional to the original signal. DSD makes use of noise shaping techniques in order to push quantisation noise up to inaudible ultrasonic frequencies. In principle, the retention of the bitstream in DSD allows the SACD player to use a basic (one bit) DAC design which incorporates a low-order analog filter. The SACD format is capable of delivering a dynamic range of 120 dB from 20 Hz to 20 kHz and an extended frequency response up to 100 kHz, although most currently available players list an upper limit of 80–90 kHz and 20 kHz is the upper limit of human hearing.
JC
altruistic.lemon said:jcbrum said:p.s. Just in case any readers don't know what DSD is, the following note might be of assistance . . .
DSD is 1-bit, has a sampling rate of 2.8224 Mhz. The output from a DSD recorder alternates between levels representing 'on' and 'off' states, and is a binary signal (called a bitstream). The long-term average of this signal is proportional to the original signal. DSD makes use of noise shaping techniques in order to push quantisation noise up to inaudible ultrasonic frequencies. In principle, the retention of the bitstream in DSD allows the SACD player to use a basic (one bit) DAC design which incorporates a low-order analog filter. The SACD format is capable of delivering a dynamic range of 120 dB from 20 Hz to 20 kHz and an extended frequency response up to 100 kHz, although most currently available players list an upper limit of 80–90 kHz and 20 kHz is the upper limit of human hearing.
JC
You seem a nice bloke, jcb, but only when you're not in patronising mode
altruistic.lemon said:JC
You seem a nice bloke, jcb, but only when you're not in patronising mode
busb said:altruistic.lemon said:Exactly the problem with this thread, then.busb said:I don't share your confidence in DB ABX testing - it doesn't deal with false negatives & needs to be conducted by professionals, is time consuming & rather boring to take part in so say nothing of the expense. No one in the industry is going to risk such tests, IMO. I've bee to demos of speaker cables where I've stated tha I couldn't really tell much difference to be challenged by the manufacturer with such statements as "Are you sure!" or "Well, it's pretty obvious to the rst of us!" etc. Such tests always start at the cheap end then work upwards. I never been to a demo whee the audiance has ever been asked to guess what level a particular cable falls in their range!
Oooh, a bullet whistling past from a sniper! What bits of digital technology you choose to believe doesn't change the technology. If it's boring you, why the hell post? If I feel sorry for anyone here, it certainly ain't me.